Recent documentation has established that the Siwi language of western Egypt, unlike most other Berber languages, has two phonemic mid vowels appearing not only in Arabic loanwords but also in inherited vocabulary: /e/ and /o/. This article examines their origin. Proto-Berber originally had a single mid vowel *e, which appears to have been retained in Siwi only before word-final /n/. In all other environments the contrast between *i and *e has been neutralized, although word-finally it seems to have survived into the 19th century. Instances of /e/ in other environments are phonetically conditioned, deriving variously from *i, *ăy, or *ă in appropriate contexts. The few attestations of /o/ are irregular, but occur in environments paralleling those in which /e/ is attested synchronically. Modern Siwi mid vowels are thus mostly secondary developments; except in final /-en/, they provide no direct evidence for the reconstruction of mid vowels in earlier intermediate stages of Berber.